The OpenCurious Newsletter
Stay curious. Get the best of the open web.
New directories, researcher tools, and open-source finds — straight to your inbox. No spam, ever.
Physical AI Companies
mapped for 2026.
A curated, regularly updated directory of the companies building Physical AI: artificial intelligence that perceives, reasons about, and acts in the real world through robots, vehicles, and machines. Founders, funding, products and headquarters, all in one place.
- 31
- companies tracked
- 6
- categories
- 2026
- data refreshed
What is Physical AI?
Physical AI (also called embodied AI) is artificial intelligence that perceives, reasons about, and acts in the real, physical world through robots, vehicles, and machines. Unlike software-only AI such as chatbots and image generators, physical AI must handle perception, dexterity, balance, and real-time control under the laws of physics, and is judged by whether a task is actually completed in the environment.
The field spans humanoid robots, robot foundation models and vision-language-action (VLA) systems, autonomous vehicles, warehouse and industrial automation, and the chips, accelerators, simulation and training data that power them: the full embodied-intelligence stack. It is widely regarded as the next major frontier after generative AI.
Below you’ll find real examples of physical AI companies, from the market leaders to emerging startups, each with founders, funding, products, headquarters, and a link to the official site. Use the search and category filters to explore humanoid, embodied-AI, robotics, and autonomous-driving companies.
Top Physical AI Companies in 2026
A quick reference to some of the most notable physical AI companies, ordered by their most recent publicly reported valuation (as of ). This is a funding-based ordering, not a quality ranking. See the full, searchable list of 31 companies below.
- 1Figure AIHumanoid Robots~$39B
General-purpose humanoid robots for the workforce, with the Helix VLA model and BotQ factory.
- 2Skild AIRobot Foundation Models~$14B
Skild Brain, an 'omni-bodied' robot foundation model that can control many different robots.
- 3WayveAutonomous Vehicles~$8.6B
End-to-end embodied AI for autonomous driving, run zero-shot across 500+ cities.
- 4Physical IntelligenceRobot Foundation Models~$5.6B
The π (pi) vision-language-action foundation models for general-purpose robot control.
- 5ApptronikHumanoid Robots~$5B
Apollo, a commercial humanoid for manufacturing and logistics, partnered with Google DeepMind.
- 6Field AIRobot Foundation Models~$2B
Field Foundation Models that let robots operate in GPS-denied, unmapped environments.
- 7Agility RoboticsHumanoid Robots~$1.75–2.1B
Digit, a bipedal robot that moves totes and bins in warehouses and distribution centers.
- 8Dexterity AIIndustrial & Warehouse~$1.65B
Full-stack AI robots that pick, pack, and palletize for FedEx, UPS, and other networks.
Physical AI Startups
Much of the field’s momentum comes from young, well-funded startups founded in the last few years. Notable physical AI startups include Physical Intelligence, Skild AI, Field AI, Generalist AI, World Labs, Moonlake AI (a 2025 world-model and simulation lab), and Proception (a 2025 Y Combinator company building dexterous robotic hands). Figure AI and 1X Technologies are older but still venture-backed and scaling.
Physical AI Robotics Companies
The robotics core of physical AI is dominated by humanoid makers such as Figure AI, Tesla (Optimus), 1X Technologies, Agility Robotics, Apptronik, Boston Dynamics, and Unitree, alongside legged and mobile robotics companies like ANYbotics and Ghost Robotics.
Physical AI Examples
Concrete examples of physical AI include humanoid robots like Figure 03 and Tesla Optimus; robot foundation models and vision-language-action (VLA) systems such as Physical Intelligence’s π0 and Figure’s Helix; autonomous vehicles from Waymo, Wayve, and Nuro; and quadrupeds like Boston Dynamics’ Spot and Unitree’s Go2. Each perceives its surroundings and takes real-world action.
Categories
Humanoid Robots
10 companies · Figure AI, Tesla (Optimus), 1X Technologies…
Robot Foundation Models
7 companies · Physical Intelligence, Skild AI, World Labs…
Autonomous Vehicles
3 companies · Waymo, Wayve, Nuro
Industrial & Warehouse
5 companies · Dexterity AI, Collaborative Robotics (Cobot), Bright Machines…
Legged & Mobile Robots
2 companies · ANYbotics, Ghost Robotics
Infrastructure & Platforms
4 companies · Nvidia (Physical AI), Hugging Face (LeRobot), Proception…
Browse all companies
Search and filter by category. 31 companies, continuously updated.
31 of 31 companies
Figure AI
Humanoid RobotsGeneral-purpose humanoid robots for the workforce.
Figure builds autonomous general-purpose humanoid robots designed to perform manual labor in manufacturing, logistics, and the home. Its Helix vision-language-action (VLA) model lets the robots see, reason, and act from natural-language instructions. Its BotQ factory, opened in 2025, is built to produce up to 12,000 humanoids per year, with a stated goal of 100,000 robots over four years.
- Founded
- 2022
- HQ
- Sunnyvale, California, USA
- Founders
- Brett Adcock
- Products
- Figure 01, Figure 02, Figure 03, Helix (VLA model), BotQ (factory)
- Funding
- Raised $1B+ in a Series C (Sept 2025) at a $39B post-money valuation, roughly 15x its prior round, bringing total funding above $1.75B. Investors include Parkway Venture Capital (lead), Brookfield, Nvidia, Intel Capital, LG, Salesforce, Qualcomm, and T-Mobile Ventures. Partners include BMW.
Tesla (Optimus)
Humanoid RobotsOptimus, a humanoid robot built on Tesla's autonomy stack.
Tesla's Optimus (Tesla Bot) is a general-purpose humanoid robot that reuses the company's vehicle AI, batteries, and actuators. Tesla aims to deploy Optimus in its own factories first, then sell it broadly, positioning it as a mass-manufactured labor platform. Its Gen 3 (V3) design is being revealed in 2026, with Fremont production lines being converted to humanoid manufacturing and a long-term target of up to 1 million units per year.
- Founded
- 2021 (program announced)
- HQ
- Austin, Texas, USA
- Founders
- Elon Musk (Tesla)
- Products
- Optimus Gen 1, Optimus Gen 2, Optimus Gen 3 (in development)
- Funding
- Self-funded within Tesla (NASDAQ: TSLA); Tesla committed ~$20B in 2026 capex partly to scale Optimus. A few hundred units were built by mid-2025, below initial targets.
1X Technologies
Humanoid RobotsSafe, soft-actuated humanoids for the home.
1X (formerly Halodi Robotics) develops humanoid robots designed to work safely alongside people. Its NEO home humanoid opened consumer pre-orders in late 2025 at $20,000 (or a $499/month subscription), with US deliveries beginning in 2026; early tasks rely partly on remote human teleoperation to gather data and improve autonomy. EVE is an earlier wheeled android. 1X emphasizes tendon-driven, low-impedance actuation for human safety and is backed by the OpenAI Startup Fund.
- Founded
- 2014
- HQ
- Moss, Norway & Sunnyvale, California, USA
- Founders
- Bernt Børnich
- Products
- NEO (home humanoid), EVE
- Funding
- Raised $130M+ in venture funding, including a ~$100M Series B (2024); investors include EQT Ventures, Tiger Global, and the OpenAI Startup Fund.
Agility Robotics
Humanoid RobotsDigit, a bipedal robot for logistics work.
Agility Robotics builds Digit, a human-centric bipedal robot designed to move totes and bins in warehouses and distribution centers. The company operates RoboFab, described as a purpose-built humanoid robot factory, and has run commercial deployments with logistics operators.
- Founded
- 2015
- HQ
- Salem & Corvallis, Oregon, USA
- Founders
- Damion Shelton, Jonathan Hurst
- Products
- Digit, RoboFab (factory)
- Funding
- Raised a ~$400M Series C (2025) at a reported ~$1.75–2.1B valuation, bringing total funding to ~$641M. Investors include WP Global Partners (lead), SoftBank, DCVC, Playground Global, Amazon Industrial Innovation Fund, NVentures (Nvidia), and Sony Innovation Fund. In June 2026 Agility announced it would go public via a merger with Churchill Capital Corp XI at a ~$2.5B valuation, with more than $620M in expected gross proceeds including a Foxconn-led PIPE, listing under the ticker AGLT. The transaction remains subject to shareholder approval and SEC review and had not closed as of this page's last update.
Apptronik
Humanoid RobotsApollo, a commercial humanoid for industrial work.
Apptronik, spun out of the Human Centered Robotics Lab at UT Austin, builds Apollo, a general-purpose humanoid for manufacturing and logistics. The company has partnered with Google DeepMind on robot intelligence and with Mercedes-Benz on factory deployments.
- Founded
- 2016
- HQ
- Austin, Texas, USA
- Founders
- Jeff Cardenas, Nick Paine
- Products
- Apollo
- Funding
- Raised a $415M Series A (Feb 2025) co-led by B Capital and Google, plus a $520M extension (Feb 2026) at a ~$5B valuation, a ~$935M Series A and nearly $1B in total funding. Backers include Mercedes-Benz, Peak6, AT&T Ventures, and John Deere.
Sanctuary AI
Humanoid RobotsPhoenix, humanoids with human-like dexterity.
Sanctuary AI develops Phoenix, a humanoid robot, and Carbon, an AI control system intended to give robots human-like reasoning and dexterous hands. The company focuses on teleoperation-to-autonomy data pipelines for general-purpose work.
- Founded
- 2018
- HQ
- Vancouver, British Columbia, Canada
- Founders
- Geordie Rose, Suzanne Gildert
- Products
- Phoenix, Carbon (control system)
- Funding
- Backed by investors including Bell and the Government of Canada SIF program.
Boston Dynamics
Humanoid RobotsAtlas, Spot & Stretch, pioneers of dynamic robotics.
A robotics pioneer (now owned by Hyundai), Boston Dynamics builds the all-electric Atlas humanoid, the Spot quadruped, and the Stretch warehouse robot. It unveiled the production version of the electric Atlas at CES 2026 and is partnering with Google DeepMind to integrate foundation models; demos show Atlas using reinforcement learning to lift 100-pound loads. Hyundai plans to deploy Atlas fleets across US plants, with a factory targeting ~30,000 units per year by 2028.
- Founded
- 1992
- HQ
- Waltham, Massachusetts, USA
- Founders
- Marc Raibert
- Products
- Atlas (electric), Spot, Stretch
- Funding
- Acquired by Hyundai Motor Group (2021); backed by Hyundai's multi-billion-dollar US robotics and manufacturing investment.
Unitree Robotics
Humanoid RobotsLow-cost humanoids and quadrupeds at scale.
Unitree is a Chinese robotics maker known for aggressively priced quadrupeds and the G1 and H1 humanoid robots. Its low price points have made advanced legged robots accessible to researchers and developers worldwide.
- Founded
- 2016
- HQ
- Hangzhou, China
- Founders
- Wang Xingxing
- Products
- G1, H1, Go2 (quadruped), B2 (quadruped)
- Funding
- Backed by investors including Tencent and reported pre-IPO interest.
Fourier Intelligence
Humanoid RobotsGR-series humanoids and rehabilitation robotics.
Fourier began in rehabilitation robotics and expanded into general-purpose humanoids with its GR-1 and GR-2 robots, targeting healthcare, research, and service applications.
- Founded
- 2015
- HQ
- Shanghai, China
- Founders
- Alex Gu
- Products
- GR-1, GR-2, Rehabilitation systems
UBTech Robotics
Humanoid RobotsWalker, humanoids for industry and service.
UBTech is a publicly listed Chinese robotics company building the Walker humanoid line for industrial and consumer service roles, alongside education and consumer robots.
- Founded
- 2012
- HQ
- Shenzhen, China
- Founders
- James Zhou
- Products
- Walker S, Walker X
- Funding
- Publicly listed (HKEX: 9880).
Physical Intelligence
Robot Foundation ModelsFoundation models that bring general intelligence to robots.
Physical Intelligence (π) is building a single general-purpose AI model to control any robot for any task. Its π (pi) vision-language-action models, from π0 (pi-zero) through π0.6, are trained across diverse robots and tasks and refined with real-world reinforcement learning to fold laundry, package products, and make coffee, aiming to be a foundation model for physical actions much like LLMs are for text.
- Founded
- 2024
- HQ
- San Francisco, California, USA
- Founders
- Karol Hausman, Sergey Levine, Chelsea Finn, Brian Ichter, Lachy Groom
- Products
- π0 (pi-zero), π0.5, π0.6
- Funding
- Raised a $70M seed (2024), a $400M Series A (Nov 2024) at a $2.4B valuation, and a $600M Series B (Nov 2025, led by CapitalG) at a $5.6B valuation, ~$1.07B total. Investors include Jeff Bezos, OpenAI, Thrive, Lux, and Bond.
Skild AI
Robot Foundation ModelsA general-purpose 'brain' for robots.
Skild AI, founded by Carnegie Mellon professors, is building Skild Brain, which it describes as an 'omni-bodied' robot foundation model that can control any robot (quadrupeds, humanoids, tabletop arms, mobile manipulators) without prior knowledge of its exact body form, generalizing across tasks and environments from manipulation to locomotion.
- Founded
- 2023
- HQ
- Pittsburgh, Pennsylvania, USA
- Founders
- Deepak Pathak, Abhinav Gupta
- Products
- Skild Brain
- Funding
- Raised a $135M Series B (2025) at a $4.5B valuation, then a $1.4B Series C (Jan 2026) led by SoftBank at a $14B+ valuation. Other investors include Nvidia, Macquarie, Lightspeed, and Coatue.
World Labs
Robot Foundation ModelsLarge World Models for spatial intelligence.
Founded by 'godmother of AI' Fei-Fei Li, World Labs builds Large World Models (LWMs) that understand and generate 3D physical worlds. Its first commercial product, Marble (launched 2025), generates explorable 3D virtual worlds from a text or image prompt. Spatial intelligence is a foundational capability for embodied AI, robotics, and simulation.
- Founded
- 2024
- HQ
- San Francisco, California, USA
- Founders
- Fei-Fei Li, Justin Johnson, Christoph Lassner, Ben Mildenhall
- Products
- Marble (3D world model), Large World Models (spatial intelligence)
- Funding
- Emerged from stealth with $230M (2024), then raised ~$1B (2026). Investors include Nvidia, AMD, a16z, NEA, Autodesk, and Fidelity.
Covariant
Robot Foundation ModelsRFM-1, a foundation model for robotic manipulation.
Covariant pioneered AI-driven robotic picking for warehouses and introduced RFM-1, a robotics foundation model trained on large volumes of real-world manipulation data. In 2024, much of its founding team joined Amazon in a licensing arrangement.
- Founded
- 2017
- HQ
- Emeryville, California, USA
- Founders
- Pieter Abbeel, Peter Chen, Rocky Duan, Tianhao Zhang
- Products
- RFM-1, Covariant Brain
- Funding
- Raised $200M+; key leadership joined Amazon in 2024.
Field AI
Robot Foundation ModelsField Foundation Models for robots in the wild.
Field AI develops Field Foundation Models (FFMs) that let robots operate autonomously in complex, GPS-denied, unstructured environments such as construction sites, mines, and industrial plants, without prior maps.
- Founded
- 2023
- HQ
- Mission Viejo, California, USA
- Founders
- Ali Agha
- Products
- Field Foundation Models (FFMs)
- Funding
- Raised ~$405M across two rounds ($91M in late 2024 and $314M in Aug 2025) at a $2B valuation. Investors include Nvidia (NVentures), Bezos Expeditions, Bill Gates' Gates Frontier, Khosla Ventures, Intel Capital, Temasek, and Samsung.
Generalist AI
Robot Foundation ModelsHardware-agnostic foundation models for general-purpose robots.
Generalist AI builds large-scale, hardware-agnostic robot foundation models that give robots broad dexterity and reasoning by scaling real-world and simulated robot-learning data. Its GEN-1 model targets high-reliability, fast execution on dexterous manipulation tasks across different robot bodies. The founding team draws from OpenAI, Google DeepMind, and Boston Dynamics.
- Founded
- 2024
- HQ
- San Mateo, California, USA
- Founders
- Pete Florence, Andy Zeng, Andrew Barry
- Products
- GEN-1 (robot foundation model)
- Funding
- Raised a $400M round (June 2026) led by Radical Ventures, bringing total funding to over $500M. Investors include Nvidia's NVentures, Boldstart Ventures, Spark Capital, Bezos Expeditions, 8VC, Union Square Ventures, and Norwest.
EmbodyX
Robot Foundation ModelsAI models that perceive, understand, and act in the physical world.
EmbodyX develops embodied-AI models designed to perceive, understand, and act in dynamic real-world environments, so robots, vehicles, and cameras can operate intelligently. The company positions itself as building 'AI for the physical world.'
- HQ
- San Mateo, California, USA
Waymo
Autonomous VehiclesFully autonomous ride-hailing at scale.
Waymo, an Alphabet subsidiary, operates one of the world's leading fully driverless robotaxi services, with paid public rides across multiple US cities. Its Waymo Driver combines lidar, radar, cameras, and deep learning for full self-driving.
- Founded
- 2009 (as the Google Self-Driving Car Project)
- HQ
- Mountain View, California, USA
- Founders
- Google / Alphabet
- Products
- Waymo One (robotaxi), Waymo Driver
- Funding
- Raised multi-billion-dollar external rounds; subsidiary of Alphabet (NASDAQ: GOOGL).
Wayve
Autonomous VehiclesEmbodied AI for end-to-end autonomous driving.
Wayve pioneers 'AV2.0', an end-to-end, learning-based approach to self-driving that uses a single neural network and embodied AI rather than hand-coded rules and HD maps. It has driven zero-shot (without city-specific fine-tuning) in 500+ cities across Europe, North America, and Japan, and powers upcoming Uber robotaxi trials starting in London in 2026.
- Founded
- 2017
- HQ
- London, United Kingdom
- Founders
- Alex Kendall, Amar Shah
- Products
- Wayve AI Driver, GAIA (driving world model)
- Funding
- Raised a ~$1.05B Series C (2024, led by SoftBank) and a $1.2B Series D (Feb 2026) at an $8.6B valuation. Investors include Nvidia, Uber, Microsoft, SoftBank, and automakers Mercedes-Benz, Nissan, and Stellantis.
Nuro
Autonomous VehiclesAutonomous delivery and a licensable AI driver.
Nuro builds autonomous vehicles for local goods delivery and licenses the Nuro Driver, its full self-driving system, to automakers and mobility partners.
- Founded
- 2016
- HQ
- Mountain View, California, USA
- Founders
- Dave Ferguson, Jiajun Zhu
- Products
- Nuro Driver, Autonomous delivery vehicles
- Funding
- Raised a reported $2B+ across rounds; investors include SoftBank and T. Rowe Price.
Dexterity AI
Industrial & WarehouseAI-powered robots for warehouses and logistics.
Dexterity builds full-stack AI robots that pick, pack, palletize, and load goods in warehouses and parcel networks for customers including FedEx and UPS. Its dual-arm 'superhumanoid' Mech has a ~5.4 m working envelope and can lift up to ~60 kg (132 lb), combining force-aware manipulation with physical AI models.
- Founded
- 2017
- HQ
- Palo Alto, California, USA
- Founders
- Samuel Zapolsky
- Products
- Mech (superhumanoid), Palletization & induction systems
- Funding
- Raised a $95M round (Mar 2025) at a $1.65B valuation, ~$291M total. Investors include Lightspeed, Kleiner Perkins, and Sumitomo.
Collaborative Robotics (Cobot)
Industrial & WarehousePractical autonomous mobile robots for work.
Founded by former Amazon Robotics VP Brad Porter, Cobot builds autonomous mobile robots designed to work safely and intuitively alongside people in warehouses, hospitals, and industrial settings, prioritizing practical deployment over humanoid form factors.
- Founded
- 2022
- HQ
- Santa Clara, California, USA
- Founders
- Brad Porter
- Products
- Proxie (mobile robot)
- Funding
- Raised a reported ~$100M Series B; investors include General Catalyst and Sequoia.
Bright Machines
Industrial & WarehouseSoftware-defined microfactories.
Bright Machines combines robotics, computer vision, and machine learning into 'Microfactories', configurable, software-defined assembly cells that automate electronics manufacturing.
- Founded
- 2018
- HQ
- San Francisco, California, USA
- Founders
- Amar Hanspal, Tzahi Rodrig
- Products
- Brightware (software), Microfactories
- Funding
- Raised a reported $400M+ across rounds; investors include BlackRock and Nvidia.
Standard Bots
Industrial & WarehouseAffordable, AI-driven robotic arms.
Standard Bots builds RO1, a six-axis robotic arm with built-in AI, aiming to make industrial automation affordable and easy to program for manufacturers and small businesses.
- Founded
- 2017
- HQ
- Glen Cove, New York, USA
- Founders
- Evan Beard
- Products
- RO1 (robotic arm)
- Funding
- Backed by investors including General Catalyst.
Pickle Robot
Industrial & WarehouseAutonomous truck and container unloading.
Pickle Robot builds physical-AI systems that autonomously unload boxes from trucks and shipping containers, one of logistics' most physically demanding tasks.
- Founded
- 2018
- HQ
- Cambridge, Massachusetts, USA
- Founders
- AJ Meyer
- Products
- Pickle unload system
- Funding
- Raised a reported $50M+ Series B.
ANYbotics
Legged & Mobile RobotsANYmal, autonomous inspection for heavy industry.
An ETH Zurich spin-off, ANYbotics builds ANYmal, a four-legged robot for autonomous inspection of industrial plants, energy facilities, and offshore sites in hazardous environments.
- Founded
- 2016
- HQ
- Zurich, Switzerland
- Founders
- Péter Fankhauser, Marco Hutter, and team
- Products
- ANYmal (quadruped)
- Funding
- Raised a reported $50M+ Series B.
Ghost Robotics
Legged & Mobile RobotsRugged quadrupeds for defense and security.
Ghost Robotics builds rugged quadrupedal 'Q-UGVs' (unmanned ground vehicles) used in defense, security, and inspection across harsh, unstructured terrain.
- Founded
- 2015
- HQ
- Philadelphia, Pennsylvania, USA
- Founders
- Gavin Kenneally, Avik De
- Products
- Vision 60 (Q-UGV)
Nvidia (Physical AI)
Infrastructure & PlatformsThe compute and simulation stack powering physical AI.
Nvidia provides much of the foundational infrastructure for physical AI and aims to be the 'Android of generalist robotics': Isaac for robot development, the Cosmos world foundation models (incl. Cosmos Reason) for synthetic data and reasoning, Isaac GR00T open humanoid foundation models (N1/N1.6/N1.7, with GR00T N2 in the works), the Newton physics engine, Omniverse for simulation, and Jetson AGX Thor edge computers for on-robot inference. Adopters include AGIBOT, LG, NEURA Robotics, and Humanoid.
- Founded
- 1993
- HQ
- Santa Clara, California, USA
- Founders
- Jensen Huang, Chris Malachowsky, Curtis Priem
- Products
- Isaac GR00T (N1.7 / N2), Cosmos (world foundation models), Newton (physics engine), Omniverse, Jetson AGX Thor
- Funding
- Publicly listed (NASDAQ: NVDA); also a major investor in physical-AI startups (Figure, Skild, Field AI, Wayve, Agility, and others).
Hugging Face (LeRobot)
Infrastructure & PlatformsOpen-source robot learning and shared datasets.
Hugging Face's LeRobot project provides open-source models, datasets, and tools for real-world robot learning, lowering the barrier to training and sharing physical-AI policies, and the company has moved into affordable open robot hardware.
- Founded
- 2016
- HQ
- New York, New York, USA
- Founders
- Clément Delangue, Julien Chaumond, Thomas Wolf
- Products
- LeRobot, Open robot datasets & policies
Proception
Infrastructure & PlatformsDexterous robotic hands and tactile data for humanoids.
Proception builds dexterous robotic hands and the tactile data pipelines needed to train them, aiming to be the go-to hand supplier for humanoid and manipulation companies that don't want to build dexterity in-house. Its ProHand is a 22-degree-of-freedom, tendon-driven hand, and its ProGlove is a wearable system for collecting human hand-interaction data. The company settled a trade-secret suit with Tesla in 2026.
- Founded
- 2025
- HQ
- Mountain View, California, USA
- Founders
- Jay Li, Jack Xu
- Products
- ProHand 1.0 (22-DoF robotic hand), ProGlove (data-collection wearable)
- Funding
- Raised $11M in seed funding (June 2026) led by First Round Capital, with Y Combinator and BoxGroup participating.
Moonlake AI
Infrastructure & PlatformsWorld models that turn the real world into sim-ready environments.
Moonlake AI builds world models and simulation infrastructure for physical and embodied AI. Its Real-to-Sim API converts an image, a video, or a plain-text description into a physically accurate, simulation-ready 3D asset with joints, articulation ranges, and drive limits already defined, exporting directly to USD, NVIDIA Isaac Sim, and MuJoCo. The pitch is world generation rather than 3D asset generation: output that behaves correctly under simulation, not just output that looks right, so it can be used for robot policy training, domain randomization, digital twins, and embodied-agent evaluation without a manual rigging pass.
- Founded
- 2025
- HQ
- San Francisco, California, USA
- Founders
- Fan-Yun Sun, Sharon Lee
- Products
- Real-to-Sim API, World Modeling Stack
- Funding
- Raised $28M in seed funding (October 2025) led by AIX Ventures, Threshold, and NVIDIA Ventures, with angel participation from Ian Goodfellow, Jeff Dean, Naval Ravikant, Steve Chen, Guillermo Rauch, and Emery Wells. Named to the CB Insights AI 100 in 2026.
Frequently asked questions
What is Physical AI?
Physical AI (also called embodied AI) is artificial intelligence that perceives, reasons about, and acts in the real, physical world through robots, vehicles, and machines. Unlike software-only AI such as chatbots, physical AI must handle perception, dexterity, balance, and real-time control under the laws of physics.
How is Physical AI different from generative AI?
Generative AI produces digital outputs, text, images, code, from data it has seen. Physical AI takes intelligence into the real world: it controls actuators and sensors, must respect physics and safety, and is judged by whether a task is actually completed in the environment, not just whether an answer looks correct.
What are the leading Physical AI companies?
Leading physical AI companies include humanoid robot makers (Figure AI, valued ~$39B; Tesla Optimus; 1X; Apptronik, ~$5B; Agility Robotics; Boston Dynamics, owned by Hyundai), robot foundation-model labs (Skild AI, ~$14B; Physical Intelligence, ~$5.6B; Field AI, ~$2B; World Labs), autonomous-driving companies (Waymo, Wayve at ~$8.6B, Nuro), warehouse and industrial robotics firms (Dexterity, Collaborative Robotics), and infrastructure providers (Nvidia). OpenCurious maintains a regularly updated, categorized directory of these companies.
How big is the Physical AI market?
Estimates vary widely by definition. Narrow 'physical AI' software/platform forecasts put the market around $1.5B in 2026 growing to ~$15B by 2032 (roughly 47% CAGR), while broader definitions that include hardware value it far higher. The humanoid robot segment alone is estimated at several billion dollars in 2026 and is projected to reach $150B+ by the mid-2030s. Venture and corporate investment into humanoid and embodied-AI startups has surged into the multiple billions of dollars per year, with Goldman Sachs projecting cumulative humanoid investment to surpass $50B by 2030.
What is a robot foundation model or vision-language-action (VLA) model?
A robot foundation model is a large AI model trained across many robots and tasks so it can generalize to new situations, much like a large language model does for text. Vision-language-action (VLA) models take in images and natural-language instructions and output robot actions, examples include Figure's Helix and Physical Intelligence's π0.
Why is Physical AI important now?
Advances in foundation models, simulation, low-cost hardware, and real-world robot data have converged, making general-purpose robots commercially plausible for the first time. Physical AI promises to address labor shortages in manufacturing, logistics, and elder care, and is widely viewed as the next major frontier after generative AI.
What are examples of Physical AI?
Examples of physical AI include humanoid robots such as Figure 03 and Tesla Optimus; robot foundation models and vision-language-action (VLA) systems like Physical Intelligence's π0 and Figure's Helix; autonomous vehicles from Waymo, Wayve and Nuro; warehouse and industrial robots from Dexterity, Covariant and Standard Bots; and quadrupeds like Boston Dynamics' Spot and Unitree's Go/B-series. Each perceives its surroundings and takes real-world action, rather than only producing text or images.
What is the difference between Physical AI and embodied AI?
The terms are used interchangeably. 'Embodied AI' (or embodied intelligence) emphasizes that the intelligence lives in a physical body, a robot, vehicle, or machine, that senses and acts, while 'Physical AI' is the term popularized by Nvidia and investors for the same idea. Companies working on embodied intelligence, such as Skild AI, Physical Intelligence, Generalist AI and Field AI, are the same set catalogued in this directory.
What chips and hardware power Physical AI?
Physical AI runs on edge robotics compute such as Nvidia's Jetson Thor and the Jetson AGX line, paired with the Isaac and GR00T robot software stacks; Qualcomm and other vendors also supply low-power inference chips for on-robot perception and control. Training the underlying robot foundation models happens in the data center on GPU clusters. Nvidia is the dominant infrastructure and accelerator provider for the category, which is why it appears in this directory under Infrastructure & Platforms.
Where does the training data for Physical AI come from?
Robot foundation models are trained on a mix of teleoperation demonstrations (humans piloting robots), large-scale simulation and synthetic data (for example with Nvidia Isaac Sim), internet video, and real-world data collected by deployed robot fleets. Sourcing high-quality, diverse robot data is one of the central bottlenecks in physical AI, and several startups focus specifically on data platforms and simulation for embodied AI.
How do Physical AI startups protect their IP?
Physical AI companies protect intellectual property through patents on hardware (actuators, hands, sensors, and full robot designs), trade secrets around their robot foundation models and training data, and proprietary simulation and control software. Because much of the value sits in learned model weights and real-world data pipelines rather than any single patent, data and model advantages are often the most defensible moat.
Which are the top physical AI companies in 2026?
Ordered by most recent publicly reported valuation, notable physical AI companies in 2026 include Figure AI (~$39B), Skild AI (~$14B), Wayve (~$8.6B), Physical Intelligence (~$5.6B), Apptronik (~$5B), Field AI (~$2B), Agility Robotics (~$1.75-2.1B), and Dexterity AI (~$1.65B), spanning humanoid robots, robot foundation models, autonomous driving, and warehouse robotics. This is a funding-based ordering, not a quality ranking.
What are the top 10 embodied AI companies?
Embodied AI and physical AI describe the same field, so the leading embodied AI companies are the leading physical AI companies. Ten that span the major categories in this directory are Figure AI, Tesla (Optimus), Apptronik and Boston Dynamics in humanoid robots; Physical Intelligence and Skild AI in robot foundation models; Waymo and Wayve in autonomous driving; Dexterity in industrial and warehouse robotics; and Nvidia, whose simulation and compute stack most of the others build on. Any ten-company list is a judgement call: this one weights publicly reported funding, shipped products and category coverage, and it necessarily leaves out strong companies in every group. The full directory catalogues them all across six categories - humanoid robots, robot foundation models, autonomous vehicles, industrial and warehouse robotics, legged and mobile robots, and infrastructure and platforms.
What companies are working on physical AI?
Companies working on physical AI fall into five groups. Humanoid robot makers: Figure AI, Tesla Optimus, 1X Technologies, Apptronik, Agility Robotics, Sanctuary AI, Boston Dynamics, Unitree, Fourier Intelligence and UBTech. Robot foundation-model and embodied-intelligence labs: Physical Intelligence, Skild AI, World Labs, Covariant, Field AI, Generalist AI and EmbodyX. Autonomous vehicle companies: Waymo, Wayve and Nuro. Industrial and warehouse robotics: Dexterity, Collaborative Robotics, Bright Machines, Standard Bots and Pickle Robot. Legged and mobile robots: ANYbotics and Ghost Robotics. Infrastructure, simulation and open datasets: Nvidia, Hugging Face (LeRobot), Proception and Moonlake AI. Each is profiled in this directory with founders, headquarters, flagship products and publicly reported funding.
What is the difference between a physical AI company and a physical AI data provider?
A physical AI company builds the robots, vehicles or foundation models that act in the world; a physical AI data provider is a services vendor that collects, teleoperates, or annotates the real-world and simulated data those models train on. This directory covers the first group, the companies building embodied AI systems, and does not rank data-collection or annotation vendors. The nearest equivalents it does cover sit in the infrastructure and platforms category: Hugging Face's LeRobot project, which publishes open robot datasets and policies; Proception, which builds dexterous robotic hands and tactile data for humanoids; and Moonlake AI, which works on world models and simulation. None of them are annotation or teleoperation service vendors.
About this directory: OpenCurious curates this list of physical AI and embodied AI companies for researchers, founders, operators, and the simply curious. Funding figures and valuations are approximate and based on publicly reported information as of ; they change frequently. To suggest a company or correction, email hello@opencurious.com.